LG AI Research Holds the World’s First “Zero-Shot Image Captioning” Challenge

At CVPR (Computer Vision and Pattern Recognition) 2023, LG AI Research will be hosting the “NICE: New frontiers for zero-shot for Image Captioning Evaluation Challenge & Workshop.” An open challenge that all AI researchers around the world can participate in, this challenge is about evaluating “how accurately AI understands and describes an image it sees for the first time” through “Zero-shot Image Captioning.” The final winners of this challenge will be given the opportunity to present their achievements at the upcoming CVPR 2023 workshop.

The world’s first workshop on Zero-shot Image Captioning

Zero-shot Image Captioning is a technology that, like human visual recognition abilities, allows AI to understand and make inferences based on previously learned data and explain these results via text when it sees images that contain objects, animals, and landscapes it has seen for the first time. NICE: New frontiers for Zero-shot for Image Captioning Evaluation Challenge is the world's first challenge related to such Zero-shot Image Captioning. The challenge runs from February to the end of April, and winners will receive prizes of up to $5,000 as well as the opportunity to present at the CVPR 2023 workshop.

The accuracy and fairness of image recognition AI technology will improve as Zero-shot Image Captioning technology gradually develops, further fostering the development of technology that directly benefits people in their daily lives. For instance, search convenience and accuracy will improve if AI generates captions and keywords automatically for massive image data online. What’s more, it can be used as a “medical expert AI” that analyzes medical images if more medical data is learned. Additionally, it is anticipated to bring innovation to the field of image search, much like “ChatGPT,” which recently made waves in the natural language search field.

LG AI Research to host jointly with Seoul National University and Shutterstock

LG AI Research will hold this workshop together with the “Graduate School of AI at Seoul National University,” which is currently researching super-giant multimodal AI and EXAONE, by establishing a joint research center, as well as with “Shutterstock”, which is jointly preparing commercialization services for image captioning AI. Shutterstock provides 26,000 free high-quality image-to-text datasets that have been AI-ethics tested for bias, sensationalism, and more. These datasets include not only photos, but also various types of images like illustrations and graphics, allowing contest participants to optimize their AI models and evaluate their performances without having to worry about copyrights, expenses, or quality.

Kyoung Mu Lee, an honorary chair professor at the Graduate School of AI at Seoul National University, commented, "The fact that LG AI Research, the Graduate School of AI at Seoul National University, and Shutterstock are co-hosting this challenge for the first time in the world shows that Korea's AI capabilities have already reached a world-class level, and I believe this challenge will serve as an opportunity to further enhance our international leadership."

Providing a forum for discussion at the world-renowned conference, CVPR 2023

LG AI Research will be participating in “CVPR 2023,” the world’s most prestigious conference in the field of computer vision to be held in Vancouver, Canada in June of this year, and will be hosting a workshop called “NICE: New Frontiers for Zero-shot Image Captioning Evaluation.” What’s more, the final winners of this challenge will be given the opportunity to present their achievements at this workshop.

LG AI Research plans to hold in-depth discussions on the direction and scalability of image captioning technology research and AI ethics issues with industry experts conducting AI research at Google, Microsoft, etc., as well as with global academics. Seunghwan Kim, Leader of Vision Lab at LG AI Research remarked, "LG AI Research has confirmed the possibility of making a quantum jump to not only generative AI but also to AI that recognizes its environment at a human level beyond the level of object recognition technology. Hosting a challenge on the subject of image captioning, a key, foundational technology for image understanding, at a world-class conference is an opportunity for LG to demonstrate its global presence in the field of computer vision, and with this challenge, we hope to provide a forum for discussing the meaning and necessity of research with AI researchers around the world, as well as the possibility of expansion."

This “NICE: New frontiers for zero-shot for Image Captioning Evaluation Challenge” is expected to expand the possibilities for Zero-shot Image Captioning. We await the participation of AI researchers from all around the world.

▶ CVPR 2023 NICE Workshop Official Website (Link)